What Portkey does
Portkey is an AI gateway and observability platform for unified model access, routing, retries, caching, budgets, logs, guardrails, and key management.
Portkey sits between applications and model providers. Its universal interface reduces provider-specific client code, while gateway configurations define routing, load balancing, fallbacks, retries, budgets, and access. Observability connects requests, traces, feedback, latency, and cost; virtual keys reduce direct distribution of provider credentials. The gateway is infrastructure, not the model that produces the answer.
The open-source gateway can be self-hosted. The hosted Dev plan is free with 10,000 monthly requests and short retention; Pro is currently $49 monthly for 100,000 requests and 30-day retention, with documented overage blocks; Enterprise is custom for volume, governance, support, and SaaS, hybrid, VPC, or air-gapped deployment. Model-provider charges remain separate, as can storage, egress, guardrails, and operations.
A fallback can silently change quality, safety, data region, license, price, or tool behavior. Caching can serve stale or cross-tenant content if keys are poorly designed, and detailed logs can capture prompts, personal data, secrets, or model reasoning. Portkey lists encryption and enterprise compliance controls, but teams must choose retention and deployment, redact data, segregate tenants, verify each provider, version configurations, and test failure paths.
How Portkey works
An application sends an OpenAI-style request to Portkey with a virtual key and a versioned configuration. The gateway authenticates it, applies budgets, rate limits, routing, retries, timeouts, fallbacks, caching, and configured guardrails, then translates and forwards the call to a selected hosted or local model provider. Responses, cost, latency, metadata, feedback, and traces can be recorded according to retention settings. Operators review outcomes and keep consequential tool execution in their own validated application boundary.
How to set up Portkey
Inventory providers and data paths
Map models, regions, subprocessors, credentials, sensitive fields, retention, licenses, cost, and permitted fallbacks for each use case.
Choose deployment and plan
Compare open source, Dev, Pro, and Enterprise request limits, retention, support, RBAC, SSO, hybrid, VPC, and air-gap needs.
Create scoped virtual keys
Separate environments and tenants, store provider keys centrally, apply budgets and rate limits, and rotate credentials on a schedule.
Version gateway policies
Define routes, timeouts, retries, fallbacks, cache keys, guardrails, metadata, and redaction as reviewed configurations.
Test and observe failures
Simulate provider errors, throttling, slow responses, schema drift, unsafe output, cost spikes, and fallback changes; alert and audit production traces.
Portkey FAQs
Is Portkey open source?
Its AI gateway has an open-source option. Hosted observability, collaboration, retention, governance, and enterprise deployment vary by plan.
How much does Portkey cost?
Dev is free for 10,000 monthly requests; Pro is currently $49 monthly for 100,000, while Enterprise is custom. Model usage is separate.
Does Portkey replace model providers?
No. It routes and governs calls to providers or local models; their pricing, terms, availability, and behavior still apply.
Can Portkey keep data in my environment?
Enterprise materials describe hybrid, VPC, and fully air-gapped options. Confirm the control plane, telemetry, support access, and exact data path.
Do automatic fallbacks preserve behavior?
Not necessarily. Models differ in schemas, context, safety, tool use, and price; test and approve every fallback route.
Listing reviewed 2026-07-28. Product details and pricing can change; verify important terms on the provider's website.
Related Coding AI tools
Related AI guides
Reviews
Tell the community what you made, what worked, and what you wish you knew before starting.